[llvm-branch-commits] [lld] 995539c - [LLD] [COFF] Don't add pseudo relocs for dangling references (#88487)

[Tom Stellard via llvm-branch-commits]

Author: Martin Storsjö
Date: 2024-04-16T14:40:47-07:00
New Revision: 995539ce05ba27f5ed1115c13bbe9a32a035f709

URL: https://github.com/llvm/llvm-project/commit/995539ce05ba27f5ed1115c13bbe9a32a035f709
DIFF: https://github.com/llvm/llvm-project/commit/995539ce05ba27f5ed1115c13bbe9a32a035f709.diff

LOG: [LLD] [COFF] Don't add pseudo relocs for dangling references (#88487)

When doing GC, we normally won't have dangling references, because such
a reference would keep the other section alive, keeping it from being
eliminated.

However, references within DWARF sections are ignored for the purposes
of GC (because otherwise, they would essentially keep everything alive,
defeating the point of the GC), see
c579a5b1d92a9bc2046d00ee2d427832e0f5ddec for more context.

Therefore, dangling relocations against discarded symbols are ignored
within DWARF sections (see maybeReportRelocationToDiscarded in
Chunks.cpp). Consequently, we also shouldn't create any pseudo
relocations for these cases, as we run into a null pointer dereference
when trying to generate the pseudo relocation info for it.

This fixes the downstream bug
https://github.com/mstorsjo/llvm-mingw/issues/418, fixing crashes on
combinations with -ffunction-sections, -fdata-sections,
-Wl,--gc-sections and debug info.

(cherry picked from commit 9c970d5ecd6a85188cd2b0a941fcd4d60063ef81)

Added: 
    lld/test/COFF/autoimport-gc.s

Modified: 
    lld/COFF/Chunks.cpp

Removed: 
    


################################################################################
diff  --git a/lld/COFF/Chunks.cpp b/lld/COFF/Chunks.cpp
index 39f4575031be54..e2074932bc466e 100644
--- a/lld/COFF/Chunks.cpp
+++ b/lld/COFF/Chunks.cpp
@@ -652,6 +652,13 @@ void SectionChunk::getRuntimePseudoRelocs(
         dyn_cast_or_null<Defined>(file->getSymbol(rel.SymbolTableIndex));
     if (!target || !target->isRuntimePseudoReloc)
       continue;
+    // If the target doesn't have a chunk allocated, it may be a
+    // DefinedImportData symbol which ended up unnecessary after GC.
+    // Normally we wouldn't eliminate section chunks that are referenced, but
+    // references within DWARF sections don't count for keeping section chunks
+    // alive. Thus such dangling references in DWARF sections are expected.
+    if (!target->getChunk())
+      continue;
     int sizeInBits =
         getRuntimePseudoRelocSize(rel.Type, file->ctx.config.machine);
     if (sizeInBits == 0) {

diff  --git a/lld/test/COFF/autoimport-gc.s b/lld/test/COFF/autoimport-gc.s
new file mode 100644
index 00000000000000..fef6c02eba82f9
--- /dev/null
+++ b/lld/test/COFF/autoimport-gc.s
@@ -0,0 +1,41 @@
+# REQUIRES: x86
+# RUN: split-file %s %t.dir
+
+# RUN: llvm-mc -triple=x86_64-windows-gnu %t.dir/lib.s -filetype=obj -o %t.dir/lib.obj
+# RUN: lld-link -out:%t.dir/lib.dll -dll -entry:DllMainCRTStartup %t.dir/lib.obj -lldmingw -implib:%t.dir/lib.lib
+
+# RUN: llvm-mc -triple=x86_64-windows-gnu %t.dir/main.s -filetype=obj -o %t.dir/main.obj
+# RUN: lld-link -lldmingw -out:%t.dir/main.exe -entry:main %t.dir/main.obj %t.dir/lib.lib -opt:ref -debug:dwarf
+
+#--- main.s
+    .global main
+    .section .text$main,"xr",one_only,main
+main:
+    ret
+
+    .global other
+    .section .text$other,"xr",one_only,other
+other:
+    movq .refptr.variable(%rip), %rax
+    movl (%rax), %eax
+    ret
+
+    .section .rdata$.refptr.variable,"dr",discard,.refptr.variable
+    .global .refptr.variable
+.refptr.variable:
+    .quad   variable
+
+    .section .debug_info
+    .long 1
+    .quad variable
+    .long 2
+
+#--- lib.s
+    .global variable
+    .global DllMainCRTStartup
+    .text
+DllMainCRTStartup:
+    ret
+    .data
+variable:
+    .long 42